Expressive facial style transfer for personalized memes mimic

Yanlong Tang, Xiaoguang Han, Yue Li, Liqian Ma, Ruofeng Tong

Abstract: Meme, usually represented by an image of exaggerated expressive face captioned with short text, are increasingly produced and used online to express people’s strong or subtle emotions. Meanwhile, meme mimic apps continuously appear, such as the meme filming feature in WeChat App that allow users to imitate meme expressions. Motivated by such scenarios, we focus on transferring exaggerated or unique expressions which is rarely noticed by previous works. We present a technique—“expressive style transfer”—which allows users to faithfully imitate popular memes’ unique expression styles both geometrically and textually. To conduct distortion-free transferring of exaggerated geometry, we propose a novel accurate feature curve-based face reconstruction algorithm for 3D-aware image warping. Furthermore, we propose an identity preserving blending model, based on a deep neural network, to enhance facial expressive textural details. We demonstrate the effectiveness of our method on a collection of Internet memes.